Research on Dynamic Human Motion Analysis Based on Constructing Standard Database

二十一世纪动作分析技术一直以来在计算机领域扮演着极其重要的角色, 对它研究的每一次突破都促使着人工智能、机器学习乃至医疗、娱乐媒体领域的巨大发展。而其中对动态人体动作的分析目前已取得部分成果, 如手势控制、动作预测、动作语言翻译等在科研以及市场上已有大规模应用, 但在计算机视觉领域, 动态视频相较于图像增加了时间维度, 使得对于动态图像的处理仍然有许多瓶颈。如何排除多余背景与动作以及精确识别人体动作变化这一难题仍待解决[1]。本文通过对计算机手语翻译这一项目为切入点, 研究如何整合海量标准数据成库与全新算法提高人体动态图像捕捉的精确性。

In 21st century, action recognition are always play an important part in the areas of computer science(CS).Every major breakthrough in this research all brought great development in artificial intelligence(AI), machine learning and even medical field, media.Despite there are some good results of dynamic human action recognition in gesture control, action prediction, action language translation, etc, we still have to solve many problems for dynamic image processing because dynamic images has a new attribute compared with static images, it’s called time dimension.So from this point of view, we need to research how to exclude redundant background elements and actions from an image collected by computer.Now our team try to find some important research results to improve accuracy of motion capture form a project of sign language translation.The specific method is we attempt to integrate massive data to a standard database and find a totally new algorithm to help it.